17:06 <AreaScout> Hi all, I am testing ChromiumOS on Debian, it was running quite well, but lately ChromiumOS crashes with this output https://pastebin.com/raw/Ba1mK0qD, so it's from the panfrost driver, all I was doing is to update to latest revision but going back to last working revision and the error is still the same .... I have deleted all Chrome caches, did a fresh binary install -> nothing, maybe you can give me some hints what can be wrong ? alyssa any ?

17:11 <macc24> AreaScout: "chromiumos on debian", not chromium brwoser?

17:11 <macc24> browser*

17:13 <AreaScout> macc24 nope ChromiumOS not only the browser

17:13 <macc24> AreaScout: entire chromiumos? how?

17:14 <AreaScout> entire ChromiumOS soon (services, google play store, arcvm a.s.o) via FydeOS

17:15 <AreaScout> the outcome image has the same structure like ChromeOS on Chromebook

17:15 <AreaScout> even libwidevine is included

17:15 <macc24> hmm

17:16 <AreaScout> own app store a.s.o

17:16 <macc24> on which device is it?

17:16 <AreaScout> https://store.fydeos.com/

17:16 <AreaScout> ODROID N2

17:17 <macc24> so mali g52

17:17 <macc24> AreaScout: can you bisect it?

17:17 <macc24> or even better, test with LIBGL_ALWAYS_SOFTWARE=1 in your environment

17:18 <AreaScout> fixed now !! hehe user problem ... me

17:19 <macc24> …

17:19 <AreaScout> one of the command line switches I added makes problems sporadically

17:19 <AreaScout> \o/ pfff I am happy

17:19 <AreaScout> https://www.areascout.at/Screenshot_2021-10-23_14.33.59.png

17:20 <macc24> AreaScout: can we talk on /query how did you get this working?

17:20 <AreaScout> https://forum.odroid.com/viewtopic.php?p=334946#p334946

17:21 <macc24> thank

17:22 <macc24> tho it doesn't really look like /whole/ chromiumos, more like just the ui

17:28 <AreaScout> Yes you are write, if you want to build everything go with the youtube video from FydeOS

17:30 <AreaScout> macc24 here is the ebuild file for everything https://github.com/FydeOS/chromium_os-raspberry_pi/blob/master/project-cros-pi/chromeos-base/chromeos-chrome/chromeos-chrome-92.0.4515.46_rc-r1.ebuild

17:31 <AreaScout> There is the audio server, arcvm for running Android VM and everything else

17:31 <AreaScout> and here is the video demonstration of the build process https://youtu.be/og4wzzIfGA0

17:32 <alyssa> AreaScout: I see Bifrost instructions. That's definitely my bug to fix.

17:34 <AreaScout> hi alyssa :) thank you in forward for investigating

17:34 <alyssa> The shader looks fine to me, so guessing it's a bug in the validator itself

17:34 <alyssa> Ironic, isn't it

17:35 <AreaScout> if you find time of course, sometimes I also see some of this blend errors with PPSSPP

17:35 <AreaScout> yes it's for sure ironic

17:36 <alyssa> AreaScout: Um.. there's supposed to be more output..

17:37 <alyssa> Oh. right.

17:37 <alyssa> There should be a line or two of output right above "Uninitialized data read after NIR -> BIR"

17:37 <alyssa> a number or r123 or something like that

17:38 <AreaScout> 38 ? could it be

17:39 <AreaScout> here is more https://pastebin.com/raw/EmB2t2qj

17:40 <alyssa> 38, that's the one

17:41 <alyssa> Ooh

17:41 <alyssa> My eyes glazed over that, look at that...

17:41 <alyssa> Z.v2i16 32.h00

17:41 <alyssa> br5 = MOV.i32 38[1]

17:41 <alyssa> br6 = MOV.i32 38[2]

17:41 <alyssa> 38[1] = SWZ.v2i16 32.h00

17:41 <alyssa> br4 = MOV.i32 38

17:41 <alyssa> br7 = MOV.i32 38[3]

17:41 <alyssa> that's a real bug in the compiler that the validator just caught

17:42 <AreaScout> assembler is not my strength, it was a little bit on my old c64 ^^

17:42 <alyssa> Do you mind setting `BIFROST_MESA_DEBUG=shaders` env var, rerunning, and sending the full output?

17:43 <alyssa> I want to see what's causing us to generate this buggy code.

17:43 <AreaScout> one mom

17:43 <alyssa> I'm suspicious of, like, dual source blending interacting with FP16 or something

17:44 <alyssa> will fix later, need to be doing $OTHER_WORK right now

17:44 <alyssa> thank you

17:49 <AreaScout> https://www.areascout.at/out2.txt

17:53 <alyssa> intrinsic store_output (ssa_38, ssa_0) (1, 15, 0, 144, 8396932) /* base=1 */ /* wrmask=xyzw */ /* component=0 */ /* src_type=float16 */ /* location=4 slots=1 dualsrc=1 mediump *//* fsSecondaryColorOut */

17:53 <alyssa> mediump dual source blending indeed. Ok, understood. Will fix this week.

17:54 <AreaScout> wow sounds good

17:54 <alyssa> Tempted to fix right now but don't have mesa on this computer I am supposed to be doing homework *whistles*

17:54 <AreaScout> :)

17:55 <alyssa> and see? the validator did catch a real, actual bug after all :-)

17:57 <AreaScout> good work !